Percept Voice Commands

• Outcome: Review
• Latest audit: May 28, 2026, 6:37 AM
• Version:
  1.0.0

Overview

This voice-control skill appears purpose-aligned, but it gives ambient speech a broad path to trigger OpenClaw CLI actions without enough documented safeguards.

Install only if you intentionally want voice input to control OpenClaw actions. Use distinctive wake words, avoid shared or noisy environments, confirm that high-risk commands require explicit approval, and review how to pause listening, inspect logs, and disable or remove the skill.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (3)

Vague Triggers

Medium

• Confidence: 89% confidence
• Finding: The documented default wake words include generic phrases like "take notes" and "send an email," which are likely to occur in normal conversation and can trigger unintended command capture. In a skill that forwards detected commands to an agent for execution, accidental activation increases the risk of unauthorized actions, especially in shared or noisy environments.

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: Allowing custom wake words without any specificity, exclusion criteria, or safety guidance creates an ambiguous invocation boundary. Users may configure everyday phrases that collide with ambient speech, causing unintended activations and downstream command execution through the OpenClaw agent.

Missing User Warnings

High

• Confidence: 93% confidence
• Finding: The skill description explains voice command routing and ambient speech wake-word detection, but it does not clearly warn users that speech is being continuously monitored and that matched commands may be forwarded for execution. This creates a significant consent and safety gap because users or bystanders may not understand that passive listening can directly trigger agent actions.

VirusTotal

54/54 vendors flagged this skill as clean.